Shortlisted
Aesthetics
Eryk Salvaggio
SWIM, (2024), Eryk Salvaggio. Glitched AI visuals and found footage.
SWIM is a "meditation on training data," a slowed visualization of the process through which media is broken down in generative diffusion models. These models train by removing information from images. This introduces, and then removes, digital noise over a series of steps. In "SWIM," archival footage of the dancer Nini Shipley is slowed down to nine minutes (as is the original soundtrack) as noise slowly comes to dissolve the frame. A contemplation of memory and archives, as well as how machines view the world, SWIM has been shown exhibited in Australia and screened in Europe and the United States.
Another notable element of the piece is that Shipley is swimming on top of AI-generated glitches. These glitches are a result of asking the generative AI model for images of noise. This introduces a paradox, as image generation models operate by removing noise from images in order to arrive at the image described in the prompt. By requesting "noise," the model becomes flummoxed, unable to produce noise as it is meant to identify patterns; and unable to produce fixed patterns as it is meant to create images of noise. These artifacts represent to the artist a kind of "uncomputable" state beyond the presence of archives and training data: pitching Shipley as "swimming" between the world beyond computational knowing, while dissolving into data streams nonetheless.
Shipley, Salvaggio notes, seems to float in peace in this liminal state between the unmodelable world and what is becomes the vector space.
Read more about SWIM at the artist's website, cyberneticforests.com.
